Byzantine & Eastern Music Analysis — MIR Thesis

A web-based Music Information Retrieval tool for analysing Byzantine, Greek Folk, Cypriot, and Arabic Maqam recordings. The system detects modal structure, tonic, microtonal deviations from 12-TET, and present instruments using a multi-stage DSP/ML pipeline.

Features

• Modal detection across 4 musical traditions (22 scale templates)
• Microtonal analysis — cent-level deviation from equal temperament (p10/p25/p50/p75/p90)
• Pitch contour visualisation — CREPE pitch track over time with tonic reference line
• Source separation via Demucs htdemucs (vocals / melody / bass / drums stems)
• Instrument detection — tradition-aware labels per Demucs stem energy
• Tradition selector — restricts mode templates to the correct scale family, eliminating cross-tradition confusion

Supported traditions

Tradition	Modes / Maqamat
Byzantine	Octoechos (8 modes) + Chromatic Hard/Soft + Enharmonic
Greek Folk	Greek Minor (Minore), Greek Hijaz
Cypriot	Cypriot Pentachord
Arabic Maqam	Rast, Bayati, Hijaz, Nahawand, Saba, Kurd, Ajam, Jiharkah

Tech Stack

Layer	Technology
Backend	Python 3.11, FastAPI
Pitch tracking	CREPE (CNN, TensorFlow)
Source separation	Demucs htdemucs (PyTorch)
Audio analysis	librosa, SciPy
Frontend	React 19, Vite
Visualisation	Recharts

Project Structure

Thesis/
├── backend/
│   ├── main.py          # FastAPI app & Pydantic models
│   ├── analysis.py      # Full DSP/MIR pipeline
│   └── requirements.txt
└── frontend/
    ├── src/
    │   ├── App.jsx
    │   └── components/
    │       ├── TraditionSelector.jsx
    │       ├── Uploader.jsx
    │       ├── ResultsPanel.jsx
    │       └── PitchChart.jsx
    └── package.json

Running the Project

Backend

cd backend
python -m venv .venv
.venv\Scripts\activate        # Windows
pip install -r requirements.txt
uvicorn main:app --reload

Backend runs on http://localhost:8000

Frontend

cd frontend
npm install
npm run dev

Frontend runs on http://localhost:5173 and proxies /api/* to the backend.

Analysis Pipeline

1. Demucs separates the audio into 4 stems
2. HPSS (librosa) isolates harmonic content from the full mix
3. Chroma + joint search (12 tonics × N templates, ending-weighted) detects mode & tonic
4. CREPE extracts cent-accurate pitch from the dominant stem (vocal or melodic instrument)
5. Refinement passes resolve ambiguous cases:
   • Byzantine kyrios/plagal disambiguation (absolute pitch, finalis-anchored)
   • Arabic neutral-interval disambiguation (Kurd/Bayati, Ajam/Rast)
6. Microtonal deviation statistics computed from CREPE pitch distribution
7. Pitch contour downsampled and returned for frontend visualisation

Academic References

• Chrysanthos of Madytos, Theoretikon Mega (1832) — 72-moira Byzantine interval system
• d'Erlanger, La Musique Arabe (1930–1959) — Arabic maqam corpus
• Kim et al., CREPE: A Convolutional Representation for Pitch Estimation (ICASSP 2018)
• Défossez et al., Hybrid Transformers for Music Source Separation (ISMIR 2022)
• Serra et al., compIAM — Computational analysis of Hindustani, Carnatic & Arab-Andalusian music